Loading quantized text encoder to CPU
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 45/100
- Issue 类型
- 功能
- 描述清晰度
- 基本清楚
- 活跃度
- 冷清
- 领域
- ai, machine-learning
调研方向
Start by reviewing the linked diffusers pull request comment and locating Lens's quantized model-loading and device-transfer path. Determine whether the existing model-loading code can retain the quantized representation on CPU and move it to GPU on demand. Done means documenting or implementing that flow and verifying CPU/GPU transfers without the reported 40 GB RAM expansion; the issue does not identify specific repository files or tests.
由索引模型根据 Issue 内容生成。
描述
Is the issue I have noted here https://github.com/huggingface/diffusers/pull/13837#issuecomment-4614858152 correct:
GPT-OSS is a 20B model that comes pre-quantized to mxfp4 for most of its parameters. If you load it to CPU using the Microsoft/transformer upstream code, it gets automatically dequantized, using 40 GB of RAM - but then it also requires 40 GB of VRAM when you move it to GPU.
If you load it to GPU, it materializes on GPU as 10 GB of VRAM using the kernels library - but it cannot be moved to CPU then.
It doesn't seem to be possible to load the model quantized to CPU, and move it on demand.
How do you handle this in this PR? diffusers might need some new infrastructure for this case, otherwise the vram-saving optimizations that diffusers have will fail.
I have settled for on-demand loading and discard-after-use for now, to avoid the 40 GB ram.
Would it be possible to load the quantized text model to CPU and just move it to GPU and back?
Many software packages do or a allow this with other models for consumer hardware, including Comfy, diffusers and OneTrainer
- 主要语言
- Python
- 星标
- 2
- 派生
- 0
- PR 合并指标
- 30 天内没有已合并 PR
环境准备
这个项目没有提供开发容器、Dockerfile 或贡献指南，环境需要你自己搭建：先看它的 README，通用步骤见我们的新手贡献指南。
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
microsoft/Lens 的其他 Issue
-
难度 2/5 1-3 小时 新手友好度 68/100
-
难度 3/5 1-2 天 新手友好度 12/100
-
难度 4/5 3-5 天 新手友好度 30/100
-
难度 4/5 3-5 天 新手友好度 42/100
-
难度 3/5 1-2 天 新手友好度 45/100
相似的 Issue
-
enhancement
难度 2/5 1-3 小时 新手友好度 72/100
fossasia/visdom-dev#77 ·
维护者通常 4 天内回复
-
难度 2/5 1-3 小时 新手友好度 82/100
aws-cloudformation/cfn-lint#4890 ·
维护者通常 2 天内回复
-
难度 2/5 1-3 小时 新手友好度 82/100
volcengine/OpenViking#6521 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 84/100
-
bug
难度 2/5 1-3 小时 新手友好度 77/100
AstrBotDevs/AstrBot#10720 ·
维护者通常 1 天内回复